Urban traffic demand forecasting method and device, electronic device and storage medium

Abstract

This application proposes a method and device, electronic device, and storage medium for predicting urban traffic demand. The method includes: obtaining a city grid map, mobile phone signaling data, and city characteristic statistical data for a target city; determining a target travel volume for each grid unit based on the mobile phone signaling data and city characteristic statistical data; constructing a grid feature vector for each grid unit based on the city characteristic statistical data; for any departure grid unit and any destination grid unit, using the corresponding grid feature vector and the grid distance between the two grid units as target input parameters, respectively, inputting the pre-trained deep gravity prediction model to obtain a travel selection probability between the departure grid unit and the destination grid unit; and calculating traffic travel prediction data based on the travel selection probability and the target travel volume of the departure grid unit. This application can effectively integrate multi-source data using the deep gravity model to improve the accuracy of urban traffic demand prediction.

Description

Technical Field

[0001] The embodiments of the present application relate to, but are not limited to, the field of urban planning, and in particular to a method and device for predicting urban traffic demand, an electronic device, and a storage medium. Background Art

[0002] Urban traffic demand forecasting is a core component of urban transportation planning and management, crucial for optimizing transportation resource allocation and alleviating congestion. Traffic demand refers to the total volume of travel within a specific time and space. It exhibits complex dynamic characteristics, influenced by a variety of factors, including population distribution, land use, economic activity, and residents' travel habits.

[0003] Traditional four-stage models are often used for forecasting. However, this approach relies on large-scale travel surveys, resulting in slow data updates and high costs. Furthermore, the model structure is relatively rigid, making it difficult to capture rapid changes in demand. Some methods utilize time series or statistical regression models. While effective in specific scenarios, they often struggle to fully capture the complex interactions of multiple factors, resulting in insufficient forecast accuracy. Summary of the Invention

[0004] This application aims to solve at least one of the technical problems existing in the prior art. To this end, this application provides a method and device for predicting urban traffic demand, an electronic device, and a storage medium. This application can effectively integrate multi-source data using a deep gravity prediction model to improve the accuracy of urban traffic demand prediction.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a method for predicting urban traffic demand, the method comprising:

[0006] Obtain city grid maps, mobile phone signaling data, and city characteristic statistics for the target city;

[0007] determining a target travel volume for each grid cell in the city grid map based on the city grid map, the mobile phone signaling data, and the city characteristic statistical data;

[0008] constructing a grid feature vector for each grid cell in the city grid map based on the city feature statistical data and the city grid map;

[0009] For any one of the plurality of grid cells and any one of the destination grid cells, the corresponding grid feature vector and the grid distance between the two grid cells are respectively input as target input parameters into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a travel selection probability between the departure grid cell and the destination grid cell;

[0010] Traffic travel prediction data is calculated based on the travel selection probability and the target travel volume of the departure grid unit.

[0011] In some embodiments, determining the target travel volume for each grid cell in the city grid map based on the city grid map, the mobile phone signaling data, and the city characteristic statistical data includes:

[0012] Determining a historical benchmark travel rate for each of the grid cells based on the city grid map and the mobile phone signaling data;

[0013] Extracting a plurality of grid attribute variables of each grid unit from the city characteristic statistical data; wherein the plurality of grid attribute variables include grid accessibility, land use mix, road traffic level, job-housing relationship coefficient, and grid housing price;

[0014] Inputting the historical benchmark travel rate and the plurality of grid attribute variables into a pre-built travel rate regression model so that the travel rate regression model outputs a target travel rate;

[0015] Obtain population distribution data of each of the grid units, and determine a target travel volume of each of the grid units based on the target travel rate and the population distribution data.

[0016] In some embodiments, constructing a grid feature vector for each grid cell in the city grid map based on the city feature statistical data and the city grid map includes:

[0017] For each of the grid units, the city characteristic statistical data is analyzed to obtain the population characteristics, land use characteristics, transportation convenience characteristics, and industrial facility characteristics corresponding to the grid unit;

[0018] A grid feature vector of each grid unit is constructed according to the population characteristics, the land use characteristics, the traffic convenience characteristics, and the industrial facility characteristics.

[0019] In some embodiments, for any one of the plurality of grid units and any one of the destination grid units, the corresponding grid feature vector and the grid distance between the two grid units are respectively used as target input parameters and input into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a travel selection probability between the departure grid unit and the destination grid unit, including:

[0020] extracting the grid feature vectors of the departure grid unit and the destination grid unit as a first input parameter and a second input parameter respectively;

[0021] Calculating the grid distance between the departure grid unit and the destination grid unit, and using the grid distance as a third input parameter;

[0022] inputting the first input parameter, the second input parameter, and the third input parameter as the target input parameter into an input layer of the depth gravity prediction model, so that the depth gravity prediction model performs feature learning interaction processing on the first input parameter, the second input parameter, and the third input parameter through a neural network to generate an internal representation of the interaction relationship between the departure grid unit and the destination grid unit;

[0023] The deep gravity prediction model calculates and outputs the travel selection probability through the activation function of the output layer.

[0024] In some embodiments, the pre-training process of the depth gravity prediction model includes:

[0025] Constructing a training data set based on the mobile phone signaling data, the training data set comprising a plurality of training samples, each of the training samples comprising: a departure grid unit feature vector and a destination grid unit feature vector corresponding to the first input parameter and the second input parameter, respectively; a grid distance corresponding to the third input parameter; and a target label, determined based on the mobile phone signaling data, representing an actual travel relationship between the departure grid unit and the destination grid unit;

[0026] Inputting the departure grid cell feature vector, the destination grid cell feature vector, and the grid distance in the training sample into the depth gravity prediction model to obtain a model prediction result;

[0027] Calculating the prediction error using a predefined loss function based on the model prediction result and the target label;

[0028] Based on the prediction error, the model parameters of the depth gravity prediction model are iteratively updated according to a preset optimization algorithm until a preset convergence condition is met.

[0029] In some embodiments, the calculating and obtaining the traffic travel prediction data based on the travel selection probability and the target travel volume of the departure grid unit includes:

[0030] For any departure grid unit and any destination grid unit in the city grid map, the target travel volume of the departure grid unit is multiplied by the travel selection probability between the departure grid unit and the destination grid unit to obtain the predicted travel volume between the departure grid unit and the destination grid unit.

[0031] In some embodiments, the method further comprises:

[0032] Dividing the mobile phone signaling data and the city characteristic statistical data according to preset population categories and preset time periods to obtain classified mobile phone signaling data and classified city characteristic statistical data corresponding to each population category in each time period;

[0033] Determining the classified target travel volume for each group category in each time period of each grid unit based on the city grid map, the classified mobile phone signaling data, and the classified city characteristic statistical data;

[0034] Constructing a classification grid feature vector for each group of people in each grid unit and each time period in the city grid map based on the classification city feature statistical data and the city grid map;

[0035] For any one of the plurality of grid cells and any one of the destination grid cells, the corresponding classification grid feature vector and the grid distance between the two grid cells are respectively used as target input parameters and input into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a classified travel selection probability between the departure grid cell and the destination grid cell;

[0036] Based on the classified travel selection probability and the classified target travel volume of the departure grid unit, classified traffic travel prediction data for each group category in each time period of each grid unit is calculated.

[0037] In a second aspect, an embodiment of the present application provides an urban traffic demand forecasting device, comprising:

[0038] Acquisition module, which obtains the city grid map, mobile phone signaling data and city characteristic statistics of the target city;

[0039] a determination module, which determines a target travel volume for each grid cell in the city grid map based on the city grid map, the mobile phone signaling data, and the city characteristic statistical data;

[0040] A construction module, which constructs a grid feature vector of each grid cell in the city grid map according to the city feature statistical data and the city grid map;

[0041] a prediction module, for any one of the plurality of grid cells and any one of the destination grid cells, using the corresponding grid feature vector and the grid distance between the two grid cells as target input parameters, respectively, and inputting them into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a travel selection probability between the departure grid cell and the destination grid cell;

[0042] A calculation module calculates traffic travel prediction data based on the travel selection probability and the target travel volume of the departure grid unit.

[0043] In a third aspect, an embodiment of the present application provides an electronic device comprising: a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the urban traffic demand forecasting method as described in any one of the embodiments of the first aspect of the present application.

[0044] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the storage medium stores a program, and the program is executed by a processor to implement the urban traffic demand forecasting method as described in any one of the embodiments of the first aspect of the present application.

[0045] The urban traffic demand prediction method proposed in an embodiment of the present application includes: obtaining a city grid map, mobile phone signaling data, and city characteristic statistical data of a target city; determining a target travel volume for each grid unit in the city grid map based on the city grid map, mobile phone signaling data, and city characteristic statistical data; constructing a grid feature vector for each grid unit in the city grid map based on the city characteristic statistical data and the city grid map; for any starting grid unit and any destination grid unit among a plurality of grid units, using the corresponding grid feature vector and the grid distance between the two grid units as target input parameters, respectively, and inputting them into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a travel selection probability between the starting grid unit and the destination grid unit; and calculating traffic travel prediction data based on the travel selection probability and the target travel volume of the starting grid unit.

[0046] The urban traffic demand forecasting method proposed in this application first obtains the target city's urban grid map, mobile phone signaling data, and urban characteristic statistics. This constructs a multidimensional data foundation that includes actual travel behavior reflected by mobile phone signaling, spatial geographic information reflected by the grid map, and regional socioeconomic attributes reflected by urban characteristic statistics. This overcomes the bias or bias that may arise from relying on a single data source and provides comprehensive data support for subsequent accurate forecasting. Then, the target travel volume for each grid cell is independently determined, accurately simulating the total scale of travel generation and attraction in each area of ​​the city, providing a key benchmark for subsequent travel distribution forecasting. Next, a grid feature vector is constructed for each grid cell, reflecting multidimensional information such as its functions and facilities. Combined with the physical distance between grid cells, this structured spatial and attribute information is input into a pre-trained deep gravity prediction model. This deep learning model can learn and express complex, nonlinear spatial interactions, thereby calculating the precise probability of travel choice between any two grid cells, effectively capturing the inherent laws of residents' travel distribution choices. Finally, the target travel volume for each departure grid cell, representing the "total travel volume" calculated independently in the first two steps, is combined with the travel selection probability representing the "distribution tendency" to calculate specific, quantified inter-grid traffic travel forecast data, achieving a complete and refined forecast from total travel volume to travel distribution. In summary, this application can effectively integrate multi-source data using the deep gravity prediction model to improve the accuracy of urban traffic demand forecasts.

[0047] Other features and advantages of the present application will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present application. The purposes and other advantages of the present application can be realized and obtained through the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a method for predicting urban traffic demand provided by an embodiment of the present application;

[0049] Figure 2 This is a flow chart of a method for predicting urban traffic demand provided by another embodiment of the present application;

[0050] Figure 3 This is a flow chart of a method for predicting urban traffic demand provided by another embodiment of the present application;

[0051] Figure 4 This is a flow chart of a method for predicting urban traffic demand provided by another embodiment of the present application;

[0052] Figure 5This is a flow chart of a method for predicting urban traffic demand provided by another embodiment of the present application;

[0053] Figure 6 This is a flow chart of a method for predicting urban traffic demand provided by another embodiment of the present application;

[0054] Figure 7 is a schematic diagram of an urban traffic demand prediction device provided by an embodiment of the present application;

[0055] Figure 8 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0057] It should be noted that although the device schematics illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the device or the sequence in the flowcharts. The terms "first," "second," and so on, in the specification, claims, and drawings, are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0059] With the accelerating pace of urbanization, the continued expansion of urban populations and the increasing complexity of urban spatial structures, residents' travel needs are becoming more diverse and dynamic. Urban traffic demand forecasting, as a crucial foundation for urban traffic planning, traffic management, and infrastructure development, is crucial for optimizing traffic resource allocation, improving traffic operation efficiency, alleviating traffic congestion, and promoting sustainable urban development.

[0060] Traditional urban traffic demand forecasting methods primarily include four-stage approaches (i.e., trip generation and attraction, trip distribution, mode splitting, and route allocation). These methods typically rely on large-scale resident travel survey data. Through statistical analysis and parameter calibration, they establish models that link travel demand with factors such as population, land use, and economic activity. However, these traditional methods suffer from several drawbacks: They rely heavily on periodic, large-scale travel surveys, resulting in long data collection cycles and high costs. This makes it difficult to timely reflect the dynamic changes in urban traffic demand, limiting the timeliness and accuracy of forecast results. Furthermore, the relatively rigid structure of these models makes them inflexible in adapting to the rapid changes in urban spatial structure and resident travel behavior. With the advancement of information technology, a large number of new data sources have emerged in the urban transportation field, such as mobile phone signaling and GPS trajectory data. Traditional methods struggle to effectively integrate multi-source, heterogeneous data and are unable to fully mine and utilize the rich information contained within these data.

[0061] In recent years, with the rapid development of big data and artificial intelligence technologies, machine learning-based traffic demand forecasting methods have gradually emerged. These methods can automatically learn complex nonlinear relationships in data, improving prediction accuracy and generalization. However, existing deep learning methods still face the following challenges in the field of traffic demand forecasting: how to effectively integrate multi-source heterogeneous data to build a data foundation that comprehensively reflects urban travel characteristics; how to achieve refined predictions of travel demand at high spatial resolution; and how to combine the interpretability of traditional traffic models with the powerful expressive power of deep learning models to improve model practicality. Therefore, there is an urgent need for an urban traffic demand forecasting method that can integrate multi-source data, achieve high spatial resolution, and effectively capture the complex relationships between urban space and travel behavior to meet the practical needs of modern urban traffic management and planning.

[0062] Based on this, the embodiments of the present application provide a method and device for predicting urban traffic demand, an electronic device, and a storage medium, which can effectively integrate multi-source data using a deep gravity prediction model to improve the accuracy of urban traffic demand prediction.

[0063] The urban traffic demand prediction method and device, electronic device and storage medium provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the urban traffic demand prediction method in the embodiments of the present application is described.

[0064] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0065] It should be noted that in each specific embodiment of the present application, when it comes to the need to perform relevant processing based on data related to the user's identity or characteristics, such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of such data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0066] Figure 1 This is an optional flowchart of the urban traffic demand forecasting method provided in the embodiment of the present application. Figure 1 The method may include but is not limited to steps 101 to 105.

[0067] Step 101: Obtain a city grid map, mobile phone signaling data, and city feature statistics of a target city.

[0068] Step 102 : Determine the target travel volume for each grid cell in the city grid map based on the city grid map, mobile phone signaling data, and city characteristic statistical data.

[0069] Step 103: construct a grid feature vector for each grid cell in the city grid map based on the city feature statistical data and the city grid map.

[0070] In step 104, for any starting grid unit and any destination grid unit among the multiple grid units, the corresponding grid feature vector and the grid distance between the two grid units are used as target input parameters and input into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs the travel selection probability between the starting grid unit and the destination grid unit.

[0071] Step 105 : Calculate and obtain traffic travel prediction data based on the travel selection probability and the target travel volume of the departure grid unit.

[0072] In steps 101 to 105 illustrated in the embodiment of the present application, the city grid map, mobile phone signaling data, and city characteristic statistical data of the target city are first obtained, and a multidimensional data foundation is constructed, which includes actual travel behavior reflected by mobile phone signaling, spatial geographic information reflected by the grid map, and regional socioeconomic attributes reflected by city characteristic statistical data. This overcomes the one-sidedness or bias that may be caused by relying on a single data source and lays a comprehensive data foundation for subsequent accurate predictions. Then, the target travel volume of each grid unit is independently determined, and the total scale of travel occurrence and attraction in each area of ​​the city is accurately simulated, providing a key benchmark for subsequent travel distribution predictions. Next, by constructing a grid feature vector for each grid unit that can reflect its multidimensional information such as its functions and facilities, and combining it with the physical distance between grid units, this structured spatial and attribute information is input into a pre-trained deep gravity prediction model; this deep learning model can learn and express complex, nonlinear spatial interaction relationships, thereby calculating the precise travel choice probability between any two grid units, effectively capturing the inherent laws of residents' travel distribution choices. Finally, the target travel volume for each departure grid cell, representing the "total travel volume" calculated independently in the first two steps, is combined with the travel selection probability representing the "distribution tendency" to calculate specific, quantified inter-grid traffic travel forecast data, achieving a complete and refined forecast from total travel volume to travel distribution. In summary, the embodiments of this application can effectively integrate multi-source data using the deep gravity prediction model to improve the accuracy of urban traffic demand forecasts.

[0073] In some embodiments, step 101 is intended to collect multi-source data for the target city within a specific analysis period. One of these data sources is a "city grid map," which typically divides the city's geographic space into standardized, based on basic geographic information. The one-kilometer resolution grid standard employed in the present invention divides the urban area into one-kilometer-long grid cells with unique identifiers and coordinate ranges. These grid cells form the fundamental units for subsequent spatial positioning, data aggregation, and analysis.

[0074] The second key source of dynamic behavioral data is "mobile phone signaling data." This refers to the spatiotemporal records of interactions between user terminals and communication base stations, provided by mobile communication network operators and anonymized and privacy-desensitized. When mobile phone users are active on the mobile network, data records containing information such as timestamps, encrypted user identifiers, and the locations of connected base stations are generated. In embodiments of the present invention, the core purpose of this massive amount of mobile phone signaling data is to accurately identify and reconstruct the "traffic travel chain" of urban residents. This involves identifying the user's actual travel trajectory sequence, including the starting location, possible intermediate stops, and the destination location, as well as the corresponding trip occurrence time and duration. This allows for statistical analysis of travel volume between different origin-destination pairs (OD pairs). Furthermore, while complying with privacy regulations, this data can also be combined with non-identifying information such as the user's terminal device model and some application usage characteristics to statistically infer group attributes such as gender, age group, and possible place of residence or work, supporting more detailed classification modeling.

[0075] The third is "urban characteristic statistics," a comprehensive dataset designed to comprehensively characterize the intrinsic attributes and external environment of each grid cell. This dataset deeply integrates "basic geographic information data," such as detailed urban "road network data" (including road grade, length, topological connectivity, etc.), "Point of Interest" (POI) data reflecting the distribution of urban functions (such as the location and type of various commercial facilities, public service agencies, and transportation stations), and "Area of ​​Interest" (AOI) data (such as large residential areas, parklands, and industrial areas). It also includes traditional socioeconomic statistical indicators such as population density within each grid cell, the area proportion of different land use types (such as residential, commercial, office, industrial, and green space), and the number of jobs. These urban characteristic statistics, especially the basic geographic information elements, not only provide a basis for understanding urban spatial structure, but also, in this invention, serve as explanatory variables in the baseline travel generation model and as a key input data source for the grid feature vectors required to construct the subsequent deep gravity prediction model.

[0076] In step 102 of some embodiments, it is necessary to quantify the "trip generation" link of urban traffic based on the basic data obtained in step 101. Step 102 aims to determine the "target travel volume" starting from each grid unit of the "city grid map" within a specific time period (for example, the morning rush hour of a weekday). The "target travel volume" here usually refers to the travel generation volume in the field of traffic planning, that is, the total number of trips generated or attracted by a specific area within a unit of time. The travel volume is calculated or predicted through a certain analysis model or statistical method, comprehensively utilizing the historical travel activity intensity inferred from mobile phone signaling data, factors related to travel generation in urban characteristic statistical data (such as population size, number of jobs, land use nature, etc.), and spatial units provided by the city grid map. The target travel volume represents the potential total travel demand scale of each grid unit as a travel starting point, and is the starting point for subsequent travel distribution calculations.

[0077] See also Figure 2 In some embodiments, step 102 may include but is not limited to steps 201 to 204 .

[0078] Step 201 : Determine the historical benchmark travel rate of each grid unit based on the city grid map and mobile phone signaling data.

[0079] Step 202: extract multiple grid attribute variables of each grid unit from the city characteristic statistical data.

[0080] Step 203 : Input the historical benchmark travel rate and a plurality of grid attribute variables into a pre-built travel rate regression model, so that the travel rate regression model outputs a target travel rate.

[0081] Step 204 : Obtain population distribution data of each grid unit, and determine a target travel volume of each grid unit based on the target travel rate and the population distribution data.

[0082] In step 201 of some embodiments, to obtain a baseline reference value for travel volume in each area, it is necessary to calculate and determine a historical baseline travel rate for each grid cell in the city grid map, based on the spatial units defined by the city grid map and combined with historical time series mobile phone signaling data. This process typically involves statistically analyzing the actual travel activities of residents reflected in the mobile phone signaling data. For example, the number of trips originating from each grid cell within a representative period in the past (such as several weekdays or weekends) is identified and divided by a baseline quantity for that grid cell (such as an estimated resident population or active population) to obtain a standardized ratio. This historical baseline travel rate reflects the average level of trip generation inherent in each grid cell based on historical observations.

[0083] In step 202 of some embodiments, in order to gain a deeper understanding of the potential factors affecting travel, it is necessary to extract and quantify "multiple grid attribute variables" for "each grid cell" from "city characteristic statistical data" covering multiple dimensions of information such as urban planning, socioeconomics, and transportation facilities. These variables are carefully selected based on economic geography theory and transportation planning practice to characterize the inherent characteristics of the grid cell, and specifically include:

[0084] Grid accessibility, which comprehensively measures the convenience of reaching other important functional areas of the city (such as employment centers, commercial areas, public service facilities, etc.) from the grid;

[0085] The land use mix degree quantifies the coexistence level of different functional land use types (such as residential, commercial, industrial, public services, green space, etc.) within the grid, reflecting the potential of the local area to meet the diverse activity needs of residents;

[0086] Road traffic level, this variable also includes the road network density, intersection density and public transportation (such as bus stops and subway stations) service coverage level within the grid;

[0087] The job-residence relationship coefficient, which measures the balance between the number of jobs and the size of the residential population within the grid;

[0088] Grid housing prices are an important indirect indicator of regional socioeconomic status, attractiveness, and resident income levels. Step 202 provides key explanatory variables for subsequent travel rate modeling, offering a multi-dimensional perspective for depicting the complex characteristics of grid cells.

[0089] In step 203 of some embodiments, the "historical baseline travel rate" obtained in step 201 and the "multiple grid attribute variables" extracted in step 202 are combined as input into a "pre-built travel rate regression model." This model is a mathematical relationship derived through parameter calibration based on historical data. It learns and quantifies the statistical correlation strength (reflected as regression coefficients) between historical travel rates and various grid attribute variables. During application, the predicted values ​​of the grid attribute variables for the target year (forecast year) are substituted into the calibrated model. Combined with the historical baseline rate, the model outputs a predictive "target travel rate." This target travel rate reflects the expected per-unit population travel intensity for that grid cell under predicted urban development conditions (such as land use changes, transportation improvements, and housing price fluctuations).

[0090] In step 204 of some embodiments, it is first necessary to obtain the "population distribution data of each grid unit", which usually comes from the population spatial distribution prediction results of the population census and special surveys, and in some embodiments, the population data of different population groups (such as by age and gender) and different time periods (weekdays, weekends, holidays) can be distinguished. Then, based on the "target travel rate" for a specific population group and a specific time period output in step 203, it is multiplied with the "population distribution data" of the corresponding grid unit, the corresponding population group and the time period to determine the "target travel volume of each grid unit". This "target travel volume" represents the total number of trips expected to be generated by the specific population group in the grid unit in a specific time period under specific conditions in the target year. This step completes the conversion from travel rate to travel volume, providing a starting point for total travel volume control for subsequent travel distribution predictions.

[0091] In some embodiments, the linear regression model formula for travel rate prediction is as follows:

[0092]

[0093] is the travel rate of the a-th resident in grid r during period s in period t (i.e., target year). It is calculated by calibrating the parameters of the above regression model and substituting the regression variables of the forecast year. It is the target variable of this model. is the travel rate of residents of category a in grid r during period s in period (t-1) (i.e., base year), which is the ratio of the total number of trips of category a to the total number of people in that category in the base year, obtained by cleaning the relevant fields in the mobile phone signaling data; is the accessibility of grid r in period t. Accessibility comprehensively measures the difficulty of people from a certain area to reach other major activity area destinations to obtain facilities and services, and has a direct impact on the travel of grid residents; is the land use mix index of grid r in period t, which is calculated based on the land use status and planning data in basic geographic information data. It can reflect the diversity of internal spatial forms of blocks and the degree of mixed use of urban functional spaces. It is used to measure the degree to which the mixed degree of land use types in traffic communities meets the travel needs of residents in the area. The road traffic level takes into account the urban road network and public transportation service facilities such as buses, including the road network density within grid r in period t. , road intersection density , the number of bus stops in grid r , obtained through spatial statistics of transportation network and station data in basic geographic information data; is the job-residence relationship coefficient of grid r in period t, which is derived from mobile phone signaling data and regional statistical yearbooks and can be calculated from the socioeconomic attribute database; represents the housing price in grid r during period t. It is derived from open-source housing platform data and can be accessed from a socioeconomic attribute database. Housing prices reflect the income characteristics of residents and employees in a neighborhood, as well as their likely travel patterns, and have a direct impact on the travel patterns of grid residents. - The coefficient of the regression model is obtained by substituting the base year travel rate data and various regression variables into the travel rate model and calibrating it, and then applying it to the prediction of the target year travel rate. And the population distribution data of each grid unit are multiplied together to get the target travel volume , which represents the target travel volume of the a-th group of people between grids i and j within time period s at time t.

[0094] Through steps 201 to 204, the embodiment of the present application first uses mobile phone signaling data to capture the historical benchmark travel rate of historical travel. Then, by introducing multiple grid attribute variables (such as grid accessibility, land use mix, etc.), the urban built environment and socioeconomic factors that affect travel are taken into consideration, thereby improving the model's perception and explanatory power of regional differences. Then, the travel rate regression model is used. This model can integrate historical trends and multi-dimensional current characteristics in a data-driven manner, learn the linear relationship between them, and thus make more accurate corrections and predictions on the travel rate to obtain the target travel rate. Finally, the conversion from travel rate to travel volume is completed by combining population distribution data. The synergistic effect of this series of steps makes the final output target travel volume close to reality, significantly improving the accuracy and reliability of the entire urban traffic demand forecasting method.

[0095] In step 103 of some embodiments, it is necessary to prepare for the input of the deep learning model and convert the complex urban features of each grid unit into a format that can be processed by the model. This step constructs a "grid feature vector" for each grid unit based on the "urban feature statistical data" and "city grid map" obtained in step 101. The so-called "grid feature vector" refers to an ordered list (vector) containing multiple numerical elements, where each element represents the characteristic measurement of the grid unit in a specific dimension. For example, the vector may contain quantified characteristic values ​​such as the population of the grid, the number of POIs of different types, the land use mixture index, the road network density, and the public transportation accessibility index. By aggregating and organizing the urban feature statistical data according to the spatial units of the city grid map, a feature vector that can comprehensively and structuredly describe the intrinsic properties of each grid unit is finally formed. These feature vectors will serve as key inputs for the subsequent deep learning model to understand the function and attractiveness of the grid unit.

[0096] See also Figure 3 In some embodiments, step 103 may include, but is not limited to, steps 301 to 302.

[0097] Step 301 : For each grid unit, analyze the urban characteristic statistical data to obtain the population characteristics, land use characteristics, traffic convenience characteristics, and industrial facility characteristics corresponding to the grid unit.

[0098] Step 302 : constructing a grid feature vector for each grid unit based on population characteristics, land use characteristics, transportation convenience characteristics, and industrial facility characteristics.

[0099] In step 301 of some embodiments, in order to fully and meticulously grasp the intrinsic properties of the basic units that constitute the urban texture, it is necessary to analyze and extract information from the pre-collected and organized urban feature statistical data for each grid unit in the urban grid map. This process aims to filter and summarize indicators from a vast amount of urban information to accurately reflect the status and characteristics of a specific grid unit across multiple key dimensions. Specifically, this data is organized into four categories: demographic characteristics, which describe the number of residents within the grid, their age and gender structure, employment status, and the relationship between work and residence; land use characteristics, which reveal the nature, area proportion, and spatial mix of various land uses within the grid (such as residential, industrial, commercial, public service, transportation, and green space); transportation convenience characteristics, which quantify the grid unit's geographical location advantages, the development of its internal and external road networks (including highways, expressways, trunk roads, secondary trunk roads, and branch roads), and the coverage of public transportation (bus stops, subway stations); and industrial facility characteristics, which reflect the economic vitality (such as real estate prices), commercial prosperity, and social service support within the grid. These characteristics are typically measured by the number of points of interest (POIs) and the area of ​​areas of interest (AOIs) to measure the density and scale of various facilities (such as catering, health care, education, commerce, entertainment, office space, services, and transportation ancillary facilities).

[0100] In step 302 of some embodiments, feature information from different dimensions and with diversity is integrated and formalized. The specific operation is to construct a grid feature vector for each grid unit based on these extracted feature indicators and their corresponding values. The grid feature vector here refers to an ordered list or array of values. Each element in the vector corresponds to a specific regional feature indicator value. For example, the regional feature variable system shown in Tables 1 and 2 contains 4 categories and a total of 30 specific indicators. Then the constructed grid feature vector will sequentially include a series of quantitative values ​​such as "total permanent population", "residential land area", "location accessibility", "real estate price", and "number of POIs and AOI area" of various facilities. The core purpose of constructing the grid feature vector is to generate a feature expression for computer processing for each grid unit, so that it can be used as a unified input unit and effectively applied to the subsequent machine learning model.

[0101] Table 1 Population and land use regional characteristic indicators of different regional characteristic categories

[0102]

[0103] Table 2 Transportation and industry characteristic indicators of different regional characteristic categories

[0104]

[0105] In step 104 of some embodiments, a pre-trained "deep gravity prediction model" can be used to simulate travelers' choice behavior between different destinations. The "deep gravity prediction model" here is an advanced prediction model that draws on the concepts of traditional gravity models but employs a deep learning network structure. This model is pre-trained, meaning it has learned the complex nonlinear relationships in urban travel patterns using a large amount of historical travel data (typically derived from mobile phone signaling data). In this step, for any pair of grid cells (one serving as the "departure grid cell" and the other as the "destination grid cell"), the "grid feature vector" constructed in step 103 and the "grid distance" between them are used as input parameters for the deep gravity prediction model. The neural network within the model performs complex nonlinear transformations, interactions, and learning on the input feature vectors, ultimately outputting a scalar value representing the relative likelihood that a traveler departing from the departure grid cell will choose the destination grid cell, namely, the "trip choice probability." This probability reflects the choice tendency based on the combined effects of the destination's attractiveness, starting point characteristics, and travel impedance.

[0106] See also Figure 4 In some embodiments, step 104 may include, but is not limited to, steps 401 to 404 .

[0107] Step 401 : extracting grid feature vectors of a starting grid unit and a destination grid unit as a first input parameter and a second input parameter respectively.

[0108] Step 402 : Calculate the grid distance between the starting grid unit and the destination grid unit, and use the grid distance as a third input parameter.

[0109] In step 403, the first input parameter, the second input parameter, and the third input parameter are input as target input parameters into the input layer of the deep gravity prediction model, so that the deep gravity prediction model performs feature learning interaction processing on the first input parameter, the second input parameter, and the third input parameter through a neural network to generate an internal representation of the interaction relationship between the starting grid unit and the destination grid unit.

[0110] In step 404, the deep gravity prediction model calculates and outputs the travel selection probability through the activation function of the output layer.

[0111] In step 401 of some embodiments, first, grid feature vectors representing intrinsic properties of the starting grid cells are extracted, and the vectors are designated as first input parameters; at the same time, grid feature vectors representing intrinsic properties of the destination grid cells are extracted, and the vectors are designated as second input parameters.

[0112] In step 402 of some embodiments, the spatial or temporal distance between the departure grid cell and the destination grid cell determined in step 401 is calculated. This quantified distance value, i.e., the grid distance, is determined as a third input parameter. This grid distance, as a key factor in measuring travel impedance, directly reflects the physical or temporal cost of traveling between two grid cells.

[0113] In step 403 of some embodiments, the first input parameter (departure grid feature vector), the second input parameter (destination grid feature vector), and the third input parameter (grid distance) calculated in step 402 are collectively used as target input parameters and fed into the input layer of a pre-built deep gravity prediction model. The deep gravity prediction model then utilizes its internal neural network structure to perform deep feature learning and interaction processing on these three input parameters. This process aims to discover and learn the complex, nonlinear interactions between the characteristics of the departure point, the characteristics of the destination, and the distance between them, thereby generating an internal representation that characterizes the intensity of travel interaction between a specific departure grid cell and the destination grid cell.

[0114] In step 404 of some embodiments, the deep gravity prediction model passes the internal representation generated in step 403 to its output layer and applies a pre-defined activation function (e.g., a softmax function) to transform the internal representation. Through the activation function, the model ultimately outputs a specific numerical value, which is the predicted probability of taking a trip from the departure grid cell to the destination grid cell. This probability quantitatively expresses the likelihood of a trip occurring given the characteristics of the departure and destination locations and the distance between them.

[0115] Through steps 401 to 404, the embodiment of the present application can achieve a refined, data-driven prediction of the travel choice probability. First, by extracting the specific grid feature vectors of the departure and destination (step 401) and the grid distance between them (step 402), it is ensured that the model input contains the key micro factors and impedance factors that affect travel. Then, the neural network of the deep gravity prediction model is used for feature learning and interaction processing (step 403), capturing the complex nonlinear dependencies between various factors and generating a deep understanding of the OD interaction relationship. Finally, the travel choice probability is calculated through the activation function of the output layer (step 404). This series of steps together constitutes an effective mechanism for simulating and predicting complex travel choice behaviors using deep learning technology, thereby improving the accuracy of the prediction.

[0116] See also Figure 5 In some embodiments, the pre-training process of the depth gravity prediction model may include, but is not limited to, steps 501 to 504 .

[0117] Step 501: Construct a training dataset based on mobile phone signaling data. The training dataset includes multiple sets of training samples. Each training sample includes: a departure grid unit feature vector and a destination grid unit feature vector corresponding to the first input parameter and the second input parameter, respectively; a grid distance corresponding to the third input parameter; and a target label determined based on the mobile phone signaling data that represents the actual travel relationship between the departure grid unit and the destination grid unit.

[0118] Step 502: Input the starting grid cell feature vector, the destination grid cell feature vector, and the grid distance in the training sample into the depth gravity prediction model to obtain the model prediction result.

[0119] Step 503: Based on the model prediction result and the target label, a predefined loss function is used to calculate the prediction error.

[0120] Step 504 : Iteratively update the model parameters of the depth gravity prediction model according to the prediction error based on a preset optimization algorithm until a preset convergence condition is met.

[0121] In step 501 of some embodiments, the collected mobile phone signaling data is converted into structured training samples. Each training sample specifically includes feature vectors representing the intrinsic properties of the departure and destination grid cells (these two corresponding to the first and second input parameters, respectively, in the subsequent model application phase), a grid distance (corresponding to the third input parameter) that quantifies the travel impedance between the two grid cells, and a target label. This target label is derived from statistical analysis of the mobile phone signaling data and is used to accurately represent the actual travel relationship between the departure and destination grid cells (e.g., whether a trip occurred, the frequency or probability of a trip, etc.).

[0122] In step 502 of some embodiments, a training sample is selected from a training dataset. The feature vectors of the departure and destination grid cells in the sample, as well as the grid distance between them, are input into the input layer of the deep gravity prediction model to be trained in a predetermined format. After receiving these inputs, the model uses its current model parameters to perform a series of calculations and transformations through its internal neural network layers, ultimately generating a model prediction result at the output layer. The neural network layer comprises 15 hidden layers, the first six of which have a dimension of 256, and the remaining layers have a dimension of 128. Within each hidden layer, a parameter matrix is ​​applied to the input variables. Neurons in the hidden layers use the Leaky Rectified Linear Unit (LeakyReLu) as the activation function. This prediction result represents the model's estimate of the travel relationship between the departure and destination grid cells corresponding to the training sample, based on the current parameter state.

[0123] In some embodiments, in step 503, after obtaining the model's prediction results, the accuracy of the prediction needs to be evaluated. To this end, the model prediction results obtained in step 502 are compared with the target labels of the training samples determined in step 501, which represent the actual situation. This comparison is quantified using a predefined loss function (e.g., a cross-entropy loss function). The value calculated by the loss function is the prediction error, which measures the degree of difference between the model prediction results and the true target labels.

[0124] In step 504 of some embodiments, the calculated prediction error will be used to guide the parameter optimization process of the model. A preset optimization algorithm, such as the gradient descent method, is used to adjust the internal parameters of the depth gravity prediction model according to the prediction error calculated in step 503. The goal of the optimization is to minimize the value of the loss function, that is, to reduce the prediction error. This "calculate error-update parameter" process will be iterated repeatedly for the samples in the training data set. With each iteration, the parameters of the model will be fine-tuned in the direction of reducing the overall prediction error. This iterative process will continue until a preset convergence condition is met, such as the prediction error is lower than a certain threshold, the performance of the model on the validation set is no longer significantly improved, or the preset maximum number of iterations is reached.

[0125] Through steps 501 to 504, the embodiment of the present application constructs a complete pre-training process for a deep gravity prediction model based on supervised learning. First, by using real mobile phone signaling data to construct a training dataset containing rich features and actual travel labels (step 501), the data foundation of the training process is ensured to be close to reality. Next, through the forward propagation of the model (step 502), error calculation based on the loss function (step 503), and iterative parameter update using the optimization algorithm (step 504), a closed-loop learning mechanism is formed. This mechanism enables the deep gravity prediction model to learn the complex and nonlinear inherent laws between departure and destination characteristics, spatial distance, and actual travel choices from large-scale actual travel data. The resulting trained model has its internal parameters fully optimized and can accurately capture and simulate real travel decision-making behavior, thereby outputting high-precision travel choice probabilities in the actual prediction task (such as step 104).

[0126] In step 105 of some embodiments, a synthetic calculation of the final traffic demand forecast result is performed. This step is based on the intermediate results obtained in the previous steps, combines the total amount of trip generation with the probability of trip distribution, and calculates the specific OD (Origin-Destination) flow. Specifically, it is achieved by multiplying the "target travel volume" of each "departure grid unit" determined in step 102 with the "travel selection probability" of the departure grid unit to all possible "destination grid units" in step 104. By performing this calculation on all possible origin and destination pairs (OD pairs), a complete "traffic travel forecast data" can be obtained, which is usually expressed as an OD matrix. Each element in the matrix represents the predicted traffic travel volume from grid unit i to grid unit j, the traffic connection intensity between various areas within the city, and the specific flow distribution. The specific formula is expressed as follows:

[0127]

[0128]

[0129] Where, It represents the travel distribution of the a-th group of people between grids i and j within time period s at time t, which is the target variable of this step. It represents the travel choice probability between grids i and j within time period s for group a at time t, which is obtained by the deep gravity prediction model; It represents the target travel volume of the a-th group of people in the i-th grid at time t within the s-time period, It represents a series of traffic demand variables related to travel grid, crowd, and time period at time t, which is represented by Summarized based on demand.

[0130] See also Figure 6 In some embodiments, the method provided in the embodiments of the present application may also include, but not be limited to, steps 601 to 605.

[0131] Step 601 : divide the mobile phone signaling data and the city characteristic statistical data according to the preset population categories and the preset time periods, and obtain the classified mobile phone signaling data and the classified city characteristic statistical data corresponding to each population category in each time period.

[0132] Step 602 : Determine the classified target travel volume of each group category in each grid unit and each time period based on the city grid map, the classified mobile phone signaling data, and the classified city characteristic statistical data.

[0133] Step 603 : constructing a classification grid feature vector for each group category in each grid unit in each time period in the city grid map based on the classification city feature statistical data and the city grid map.

[0134] In step 604, for any starting grid unit and any destination grid unit among the multiple grid units, the corresponding classified grid feature vector and the grid distance between the two grid units are used as target input parameters and input into the pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs the classified travel selection probability between the starting grid unit and the destination grid unit.

[0135] Step 605 : Based on the classified travel selection probability and the classified target travel volume of the departure grid unit, the classified traffic travel prediction data of each group category in each grid unit and each time period is calculated.

[0136] In step 601 of some embodiments, the mobile phone signaling data and city characteristic statistical data are divided into detailed categories according to preset population categories such as gender, age group, occupation type, etc. and preset time periods such as weekdays, weekends, holidays, morning and evening rush hours, etc., so that classified mobile phone signaling data and classified city characteristic statistical data corresponding to each population category in each time period can be obtained.

[0137] In step 602 of some embodiments, the city grid map, categorized mobile phone signaling data, and categorized city characteristic statistics are combined to determine the categorized target travel volume for each grid cell, for each time period, and for each demographic. By independently modeling travel behavior for different demographic groups and time periods, the differences in travel demand for different demographic groups at different time periods can be more accurately reflected, improving the sophistication and pertinence of the prediction results.

[0138] In step 603 of some embodiments, based on the classified city feature statistics and the city grid map, a corresponding classified grid feature vector is constructed for each grid unit, each time period, and each population category. This feature vector integrates multi-dimensional information such as population characteristics, land use characteristics, transportation convenience characteristics, and industrial facility characteristics, and is customized for different population groups and time periods, thereby providing a more discriminative and representative feature description for model input.

[0139] In step 604 of some embodiments, for any pair of departure and destination grid cells, the corresponding classified grid feature vectors are extracted, and the grid distance between them is calculated. These are then fed as target input parameters into a pre-trained deep gravity prediction model. This model can output the classified trip selection probability between the departure and destination grid cells based on the characteristics of different groups of people and time periods, accurately describing the travel distribution patterns by group and time period.

[0140] In step 605 of some embodiments, by combining the classified target travel volume of the departure point with its classified travel selection probability to each destination, fine-grained prediction of OD travel volume between different groups of people, different time periods, and different spatial units is achieved.

[0141] In summary, through steps 602 to 605 above, the present embodiment can fully mine and utilize multi-source heterogeneous data to achieve refined predictions of urban traffic demand by population, time period, and spatial unit. This method not only improves the spatial and temporal resolution of traffic demand forecasts, but also enhances the ability to depict differences in travel behavior among different population groups, thereby providing more targeted decision support for application scenarios such as urban traffic planning and travel service optimization.

[0142] See also Figure 7The embodiment of the present application further provides an urban traffic demand forecasting device, which can implement the above-mentioned urban traffic demand forecasting method, including:

[0143] Acquisition module, which obtains the city grid map, mobile phone signaling data and city characteristic statistics of the target city;

[0144] A determination module determines the target travel volume for each grid cell in the city grid map based on the city grid map, mobile phone signaling data, and city characteristic statistical data;

[0145] A construction module, which constructs a grid feature vector of each grid cell in the city grid map based on the city feature statistical data and the city grid map;

[0146] The prediction module uses the corresponding grid feature vector and the grid distance between any of the multiple grid cells as target input parameters, and inputs them into a pre-trained deep gravity prediction model. The deep gravity prediction model outputs the trip selection probability between the starting grid cell and the destination grid cell.

[0147] The calculation module calculates the traffic travel prediction data based on the travel selection probability and the target travel volume of the departure grid unit.

[0148] In a third aspect, an embodiment of the present application provides an electronic device comprising: a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the urban traffic demand forecasting method as described in any one of the embodiments of the first aspect of the present application.

[0149] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the storage medium stores a program, and the program is executed by a processor to implement the urban traffic demand forecasting method as described in any one of the embodiments of the first aspect of the present application.

[0150] See also Figure 8 , Figure 8 The hardware structure of an electronic device according to another embodiment is shown. The electronic device includes:

[0151] The processor 801 may be implemented as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0152] The memory 802 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 802 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 802 and is called by the processor 801 to execute the urban traffic demand forecasting method of the embodiments of this application.

[0153] Input / output interface 803, used to implement information input and output;

[0154] Communication interface 804, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0155] Bus 805 , which transmits information between various components of the device (e.g., processor 801 , memory 802 , input / output interface 803 , and communication interface 804 );

[0156] The processor 801 , the memory 802 , the input / output interface 803 and the communication interface 804 are connected to each other in communication within the device via a bus 805 .

[0157] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned urban traffic demand prediction method.

[0158] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0159] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0160] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0161] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0162] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0163] The terms "first," "second," "third," "fourth," and the like (if any) in the specification of the present application and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in orders other than those illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0164] It should be understood that in this application, "at least one (item)" means one or more, and "more" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or plural.

[0165] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the above-mentioned units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0166] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0167] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0168] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes multiple instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store programs.

[0169] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. A method for predicting urban traffic demand, characterized in that: The method comprises: Obtain city grid maps, mobile phone signaling data, and city characteristic statistics for the target city; determining a target travel volume for each grid cell in the city grid map based on the city grid map, the mobile phone signaling data, and the city characteristic statistical data; constructing a grid feature vector for each grid cell in the city grid map based on the city feature statistical data and the city grid map; For any one of the plurality of grid cells and any one of the destination grid cells, the corresponding grid feature vector and the grid distance between the two grid cells are respectively input as target input parameters into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a travel selection probability between the departure grid cell and the destination grid cell; Calculating traffic travel prediction data based on the travel selection probability and the target travel volume of the departure grid unit; The method further includes: dividing the mobile phone signaling data and the city characteristic statistical data according to preset population categories and preset time periods, and obtaining classified mobile phone signaling data and classified city characteristic statistical data corresponding to each population category in each time period; Determining the classified target travel volume for each group category in each time period of each grid unit based on the city grid map, the classified mobile phone signaling data, and the classified city characteristic statistical data; Constructing a classification grid feature vector for each group of people in each grid unit and each time period in the city grid map based on the classification city feature statistical data and the city grid map; For any one of the plurality of grid cells and any one of the destination grid cells, the corresponding classification grid feature vector and the grid distance between the two grid cells are respectively used as target input parameters and input into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a classified travel selection probability between the departure grid cell and the destination grid cell; Based on the classified travel selection probability and the classified target travel volume of the departure grid unit, classified traffic travel prediction data for each group category in each time period of each grid unit is calculated.

2. The urban traffic demand forecasting method according to claim 1, characterized in that: Determining the target travel volume for each grid cell in the city grid map according to the city grid map, the mobile phone signaling data, and the city characteristic statistical data includes: Determining a historical benchmark travel rate for each of the grid cells based on the city grid map and the mobile phone signaling data; Extracting a plurality of grid attribute variables of each grid unit from the city characteristic statistical data; wherein the plurality of grid attribute variables include grid accessibility, land use mix, road traffic level, job-housing relationship coefficient, and grid housing price; Inputting the historical benchmark travel rate and a plurality of the grid attribute variables into a pre-built travel rate regression model so that the travel rate regression model outputs a target travel rate; Obtain population distribution data of each of the grid units, and determine a target travel volume of each of the grid units based on the target travel rate and the population distribution data.

3. The urban traffic demand forecasting method according to claim 1, characterized in that: The step of constructing a grid feature vector for each grid cell in the city grid map based on the city feature statistical data and the city grid map includes: For each of the grid units, the city characteristic statistical data is analyzed to obtain the population characteristics, land use characteristics, transportation convenience characteristics, and industrial facility characteristics corresponding to the grid unit; A grid feature vector of each grid unit is constructed according to the population characteristics, the land use characteristics, the traffic convenience characteristics, and the industrial facility characteristics.

4. The urban traffic demand forecasting method according to claim 1, characterized in that: The method includes: for any one of the plurality of grid cells and any one of the destination grid cells, using the corresponding grid feature vector and the grid distance between the two grid cells as target input parameters, and inputting the corresponding grid feature vector and the grid distance between the two grid cells into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a travel selection probability between the departure grid cell and the destination grid cell, including: extracting the grid feature vectors of the departure grid unit and the destination grid unit as a first input parameter and a second input parameter respectively; Calculating the grid distance between the departure grid unit and the destination grid unit, and using the grid distance as a third input parameter; inputting the first input parameter, the second input parameter, and the third input parameter as the target input parameter into an input layer of the depth gravity prediction model, so that the depth gravity prediction model performs feature learning interaction processing on the first input parameter, the second input parameter, and the third input parameter through a neural network to generate an internal representation of the interaction relationship between the departure grid unit and the destination grid unit; The deep gravity prediction model calculates and outputs the travel selection probability through the activation function of the output layer.

5. The urban traffic demand forecasting method according to claim 4, characterized in that: The pre-training process of the deep gravity prediction model includes: Constructing a training data set based on the mobile phone signaling data, the training data set comprising a plurality of training samples, each of the training samples comprising: a departure grid unit feature vector and a destination grid unit feature vector corresponding to the first input parameter and the second input parameter, respectively; a grid distance corresponding to the third input parameter; and a target label, determined based on the mobile phone signaling data, representing an actual travel relationship between the departure grid unit and the destination grid unit; Inputting the departure grid cell feature vector, the destination grid cell feature vector, and the grid distance in the training sample into the depth gravity prediction model to obtain a model prediction result; Calculating the prediction error using a predefined loss function based on the model prediction result and the target label; Based on the prediction error, the model parameters of the depth gravity prediction model are iteratively updated according to a preset optimization algorithm until a preset convergence condition is met.

6. The urban traffic demand forecasting method according to claim 1, characterized in that: The calculating and obtaining the traffic travel prediction data based on the travel selection probability and the target travel volume of the departure grid unit includes: For any departure grid unit and any destination grid unit in the city grid map, the target travel volume of the departure grid unit is multiplied by the travel selection probability between the departure grid unit and the destination grid unit to obtain the predicted travel volume between the departure grid unit and the destination grid unit.

7. An urban traffic demand forecasting device, characterized in that: include: Acquisition module, which obtains the city grid map, mobile phone signaling data and city characteristic statistics of the target city; a determination module, which determines a target travel volume for each grid cell in the city grid map based on the city grid map, the mobile phone signaling data, and the city characteristic statistical data; A construction module, which constructs a grid feature vector of each grid cell in the city grid map according to the city feature statistical data and the city grid map; a prediction module, for any one of the plurality of grid cells and any one of the destination grid cells, using the corresponding grid feature vector and the grid distance between the two grid cells as target input parameters, respectively, and inputting them into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a travel selection probability between the departure grid cell and the destination grid cell; The calculation module calculates the traffic travel prediction data based on the travel selection probability and the target travel volume of the departure grid unit The device is further configured to: divide the mobile phone signaling data and the city characteristic statistical data according to preset population categories and preset time periods, and obtain classified mobile phone signaling data and classified city characteristic statistical data corresponding to each population category in each time period; Determining the classified target travel volume for each group category in each time period of each grid unit based on the city grid map, the classified mobile phone signaling data, and the classified city characteristic statistical data; Constructing a classification grid feature vector for each group of people in each grid unit and each time period in the city grid map based on the classification city feature statistical data and the city grid map; For any one of the plurality of grid cells and any one of the destination grid cells, the corresponding classification grid feature vector and the grid distance between the two grid cells are respectively used as target input parameters and input into a pre-trained deep gravity prediction model, so that the deep gravity prediction model outputs a classified travel selection probability between the departure grid cell and the destination grid cell; Based on the classified travel selection probability and the classified target travel volume of the departure grid unit, classified traffic travel prediction data for each group category in each time period of each grid unit is calculated.

8. An electronic device, characterized in that: include: A memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method for predicting urban traffic demand according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that The storage medium stores a program, and the program is executed by a processor to implement the urban traffic demand prediction method according to any one of claims 1 to 6.

Images